Hi everyone! Just bought Nvidia 680/2 to replace my old Rx 560/4 and was shocked with Deep Prime perfomance. With Radeon export took about 30 seconds to process Canon 550d image, now it takes more than two minutes on nvidia. On preferences tab nvidia shown as GTX 680* (partially supported) and selected as DP processing device. OpenCL accel. is also cheked. Modifying config file this way
Close DxO PhotoLab.
Locate and edit the following file: "C:\Program Files\DxO\DxO PhotoLab 4 or 5\DxO.PhotoLab.exe.config”.
Just under the “WinMLUseGraphicQueue” setting, replace “False” with “True”.
Restart DxO PhotoLab.
shows no efffect at all.
By the way, my config is
Amd phenom II X2 550, 12 Gb DDR3 RAM, Zotac Gtx 680/2Gb with 472.98 drivers, Win 10 Pro 19044.1526, DxO Photolab 4.0.2 4437
It may have been a significant step up from your previous card but it is clearly still not enough to process DeepPRIME. I use a GTX 1050ti which is a little more powerful than your card and also has 4 gb of ram. It is about the lowest card in the nVidia lineup which will fully support DeepPRIME. And I suppose that could change for future versions that may be more demanding.
Your Nvidia driver is out of date. Current version is 511.65. Also your version of PL4 is out of date, current version is 4.3.x This could make a difference.
Perhaps those updates will make a difference. It is certainly worth a try. However, besides any other issues with regards to that card’s specifications, I believe it only has two gigabytes of memory. It just may not be up to the task. My current card, the GTX 1050 TI, is only slightly better spec’d, but has 4 GB of memory. It works fine with DeepPrime, processing 24 megapixel images in around 20 seconds.
A GTX680 is barely an upgrade over an RX560. FP32 performance is about 25% higher and that is what Deep prime needs from the GPU so less than 25% Deep prime performance increase is to be expected.
The performance decrease you are seeing must be a driver/DirectML compatibility issue or 2GB of GPU memory just not being enough.
Don’t know how much can be read into the observation, but, my GPU monitor says it is using 3.4GB of memory while PL5 processes images. 20MP or 33MP images seems not to make a difference.
I hate to advise hardware changes because I cannot guarantee the results you would get. That card works for me and several others here. I got it because it didn’t need a separate connection to the small 300 watt power supply of my almost 6-year old computer and still allowed me to process DeepPRIME.
Thanks! You confirmed my fears of upgrade)0))0)
But as I don’t have too much money to upgrade my GPU and Cpu, I think it will be a good option to use AMD APU with vega 8/11. In benchmark shown somwhere in this forum it shows perfomance about 1MP/Sec which is good enough for me. What do you think? Vega is not as good as my actual 680 in games so I wanna use them at one time.
To help you in your choice, you can consult this section:
There is a link to a Google spreadsheet where the performance of GPUs (and others, like CPUs, DPL versions 4 and 5, and so on) are compared on test images.
This topic, and the corresponding spreadsheet are very informative.
For example, switching from DPL4 to DPL5 and updating the GPU driver is probably the most efficient way to reduce DeepPRIME’s processing time at low cost. You can also guess the expected processing time for DeepPRIME based on the FP32 score of your GPU.
I don’t know what the /2 or /4 are in your naming, but isn’t the GTX 680 a waaaay older GPU than the RX 560?
Support has ended on the nvidia side for that generation, and I wouldn’t be surprised it it couldn’t be used to it’s full potential in modern applications.
The CUDA support isn’t that different I suppose, but I don’t think Deep Prime uses Cuda directly, it uses OpenCL or another way of talking to modern GPUs.
I can’t know for sure where the bottleneck is (memory or not), but my GTX 1060 6gb breezes through files with Deep Prime, so it isn’t a nvidia vs AMD thing.
I think that GPU has less OpenCL power to give then you might think, and the old Phenom CPU isn’t helping either .
.